Group Differences in Reported Jitter and Shimmer Are Partly Associated with the Harmonics-to-Noise Ratio

Jitter and shimmer are often compared between groups that also differ in the harmonics-to-noise ratio (HNR). This study examined how much of a reported group difference in these measures is associated with the HNR difference rather than with a difference in cycle-to-cycle perturbation.

Retrospective analysis of an existing database against a simultaneously recorded electroglottographic (EGG) reference.

Sustained /a/, /i/, and /u/ at normal pitch and /a/ at low and high pitch were analyzed from the Saarbrücken Voice Database (1554-1734 speakers per condition, one recording per speaker). Reference jitter was computed from glottal closure instants detected in the EGG signal; acoustic jitter, shimmer, and HNR were obtained with Praat. In Experiment 1, groups were matched on EGG-derived reference jitter and fundamental frequency (f₀) and differed in HNR. In Experiment 2, groups were matched on HNR and differed in reference jitter. In Experiment 3, groups were formed on reference jitter without matching.

In Experiment 1 (/a/, normal pitch), an HNR difference of 6.55 dB was accompanied by a reported jitter difference of -0.257% points (pp) (95% CI -0.305 to -0.208; d = -0.63), whereas the reference jitter difference was -0.079 pp (95% CI -0.167 to 0.010). The effect was present in all five conditions, persisted after adjustment for sex, age, recording type, recording level, duration, and spectral balance (-0.214 to -0.289 pp), appeared in a second jitter implementation, and increased with the HNR gap (-0.043 pp/dB; 95% CI -0.050 to -0.027). With HNR matched, 18-34% of a reference jitter difference appeared in the reported value. A synthetic control indicated an HNR-associated shimmer difference of -1.06 pp (95% CI -1.35 to -0.76).

In this database and implementation, between-group HNR differences were associated with reported jitter and shimmer differences not accounted for by the EGG reference, while reference differences were attenuated. The design was observational, and the groups also differed in recording level and spectral balance, so these findings are associations rather than demonstrations that HNR itself produces the differences. Group comparisons of perturbation measures should be interpreted together with the accompanying HNR difference.